On a content recommendations team, I noticed we were optimising primarily for click-through rate, but I felt that wasn't capturing real user satisfaction. I proposed we also track a seven-day retention metric alongside CTR, arguing that users who clicked but never returned weren't actually getting value. The PM agreed, we added the retention signal to our dashboard, and over the next quarter we saw that some high-CTR content actually correlated with lower retention. The team started factoring that in when evaluating recommendation changes.
Our content ranking team was optimising on same-session CTR, but I could show in cohort data that the highest-CTR content formats — short provocative headlines — were producing users with thirty percent lower ninety-day retention than average. The mechanism was clear: we were rewarding content that generated regret clicks. I proposed replacing CTR as the primary ranking signal with a composite that weighted a post-interaction satisfaction proxy — derived from session depth and return visit rate within forty-eight hours. The PM resisted because CTR was the metric tied to the team's OKR. I brought the cohort analysis to the DS lead and the HM, demonstrated the retention decay curve directly, and got the OKR changed before the next half. Ranking improvements were then evaluated on the composite; CTR dropped six percent and ninety-day retention improved eleven percent.